Banca de QUALIFICAÇÃO: MATHEUS GIBEKE SIQUEIRA DALMOLIN

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : MATHEUS GIBEKE SIQUEIRA DALMOLIN
DATE: 17/10/2025
TIME: 08:30
LOCAL: meet.google.com/zwk-uskp-xdn
TITLE:

Predictive Analysis of Molecular Incompatibility in Pharmaceutical Formulations Based on Data Augmentation and Transformer Models

KEY WORDS:

Pharmaceutical Incompatibility, Data Augmentation, Large Language Models (LLMs), Transformer Models

PAGES: 79
BIG AREA: Ciências Biológicas
AREA: Biotecnologia
SUMMARY:

The incompatibility between Active Pharmaceutical Ingredients (APIs) and excipients is a challenge in drug development, marked by data scarcity and class imbalance. This qualification thesis proposes a computational framework for the predictive analysis of this incompatibility, using data augmentation strategies based on Large Language Models (LLMs). The framework infers causal mechanisms from known incompatibilities and generates chemically plausible interaction hypotheses. This approach expanded the minority incompatibility class from 344 to 2,096 instances, an increase of more than 500%, correcting the initial imbalance and overcoming the limitations of statistical resampling techniques such as SMOTE. The enriched dataset was then used to train predictive models, with an emphasis on the Transformer architecture. The methodology compares the performance of classical models (e.g., XGBoost) and a domain-specific Transformer model (ChemBERTa-2) under different data treatment scenarios. The results indicate that LLM-guided augmentation improves performance compared to traditional approaches and that the specialization of ChemBERTa-2 through fine-tuning achieved an F1-Score of 94.35%. The thesis presents a pipeline that integrates knowledge generation with LLMs and Transformer models to support the development of pharmaceutical formulations.

COMMITTEE MEMBERS:
Interno - 1837240 - MARCELO AUGUSTO COSTA FERNANDES
Externo ao Programa - 2885532 - IVANOVITCH MEDEIROS DANTAS DA SILVA - UFRNExterna à Instituição - RAQUEL DE MELO BARBOSA - UGR - Univer
Externo à Instituição - SERGIO NATAN SILVA - UFCG